Code and rainfall data for: Beyond Crash History: Rainfall and Regional Effects on Road Traffic Crash Severity in Nigeria, an Explainable Machine Learning Analysis

This record contains the analysis code and the derived rainfall series supporting the article "Beyond Crash History: Rainfall and Regional Effects on Road Traffic Crash Severity in Nigeria, an Explainable Machine Learning Analysis". The study tests whether rainfall, seasonality, public holidays and geopolitical zone improve prediction of road traffic crash frequency and a fatality-weighted severity index across Nigeria's 36 states and the Federal Capital Territory (Q4 2020 to Q1 2024; 518 state-quarters), once each state's crash history is controlled. Multiple linear regression, random forest and XGBoost models are compared against naive history baselines and interpreted with SHAP. Files- rainfall_nasa_power_quarterly.csv: quarterly rainfall at each state capital, derived from NASA POWER daily data, for the 37 states and the 14 quarters analysed.- Supplementary_Code_TIP.zip: Python scripts and a README describing the run order and requirements. - rebuild_enriched.py: assembles the analysis dataset by merging crash records with rainfall, holiday, season and zone variables. - crash_pipeline_v2.py: trains, tunes and evaluates the models and produces the SHAP analysis. - robustness_checks.py: runs the bootstrap comparisons, Poisson benchmark, repeated seeds, rainfall-anomaly and leave-one-state-out checks. Crash dataThe state-level crash records come from the Federal Road Safety Corps (FRSC) and are not included in this record. See the article's Data Availability Statement for how to obtain them. Rainfall data were obtained from the NASA Langley Research Center (LaRC) POWER Project funded through the NASA Earth Science/Applied Science Program.